Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SD05

2 - Mixed-integer Nonlinear Programming Method for Chance-constrained Models with Endogenous and Exogenous Uncertainty

n SD03 North Bldg 121C Incentive Conflicts in New Product Development: Theory and Data Sponsored: Technology, Innovation Management & Entrepreneurship Sponsored Session Chair: Jochen Schlapp, University of Mannheim, Mannheim, 68163, Germany 1 - Turning the Tables: Licensing Contracts with Reciprocal Options Pascale Crama, Singapore Management University, 50 Stamford Road, Singapore, 178899, Singapore, Niyazi Taneri R&D collaborations between an innovator and a partner are often undertaken when neither can bring the product to market individuallyùmaking joint effort necessary. Either party can avoid moral hazard by acquiring their missing capability and taking sole ownership of the project. The extent of two risksùabout whether the other party’s capability will be acquired and about how well it will be implementedùdetermine the optimality of not signing an up-front contract, signing buyout contracts, buyback contracts, dual buyout-buyback contracts, and an underutilized novel reciprocal option contract. 2 - Project Selection and Success in Pharmaceutical R&D Panos Markou, Cambridge Judge Business School, Calle Maria de Molina 12, Bajo, cambridge, United Kingdom, Nektarios Oraiopoulos, Stylianos Kavadias We analyze the R&D pipelines of the fifteen largest pharmaceutical companies and examine how firms select which projects to pursue, and which ones eventually succeed, i.e., receive FDA approval. We find that firms select projects where they have prior experience, but that selection also depends on technological signals from rivals. Additionally, we find that in-licensed projects are less likely to be selected for development than in-house projects; but, Jeremy Hutchison-Krupat, University of Cambridge, Cambridge, 22901, United Kingdom, Konstantinos Stouras, Raul Chao We study a firm that chooses to employ an innovation contest only when the contest is expected to generate more value than alternative options. The population of solvers differ in their ability to generate value and they incur an opportunity cost to participate. Thus, some solvers may not find it beneficial to participate. Critically, when the firm designs the contest, and when the solvers choose to enter, the number of participants remains uncertain. Within this setting, we find that firms with sufficiently high opportunity cost maximize expected profit through the provision of multiple awards; otherwise, firms maximize their profit through a winner-take-all award structure. 4 - Search Under Constraints Sezer Ulku, Georgetown University, 594 Rafik B. Hariri Building, Washington, DC, 20057, United States In innovation contexts, slack resources are required to allow experimentation in the face of uncertainty. At the same time, it is also suggested that “necessity is the mother of inventionö, and that constraints result in superior innovation performance. We conduct several experiments to investigate how constraints influence search behavior and the performance achieved in problem solving tasks. We find that solution quality attained under a moderate constraint is superior to that attained when no such constraint is present. As expected, when the resource constraint becomes very tight, performance suffers. n SD04 North Bldg 122A Joint Session:Integer Programming/Opt under Uncertainty: Stochastic Mixed-Integer Programming Sponsored: Optimization/Integer and Discrete Optimization Sponsored Session Chair: Amy Burton, Clemson University, Clemson, SC, United States 1 - An Affine Bounding Method for Two-stage Stochastic Integer Programs Gustavo Angulo, Macul, Santiago, 7820436, Chile, Merve Bodur, Diego A. Moran For two-stage stochastic programs with mixed-integer recourse, we propose a decomposition method akin to Benders’ decomposition. To approximate the second-stage value function, we iteratively partition the first-stage feasible set with affine lower-bounding functions derived from a Lagrangian relaxation, which is shown to be exact at the vertices of each element of the partition. Preliminary computational results are also presented. conditional on selection, they have higher likelihood of success. 3 - The Role of Participation in Innovation Contests

Alan Delgado de Oliveira, The George Washington University, Washington, DC, United States, Miguel Lejeune, Francois Margot We propose a chance-constrained stochastic programming model with endogenous and exogenous uncertainty and develop an MILP solution method. This study is motivated by the need to quickly evacuate severe casualties from the battlefield. Computational results will be presented. 3 - On Distributionally Robust Chance Constrained Program with Wasserstein Distance Weijun Xie, Virginia Tech, Blacksburg, VA, 24060, United States In a distributionally robust chance constrained optimization problem (DRCCP), the chance constraint is required to hold for all probability distributions of the uncertain parameters within a chosen Wasserstein distance from an empirical distribution. In this work, we investigate equivalent reformulations and approximations of such problems. We first show that a DRCCP can be reformulated as a conditional-value-at-risk constrained optimization problem, and thus admits tight inner and outer approximations. Next, we show that a DRCCP is mixed integer representable. We further identify submodular substructure in DRCCP and hence are able to derive valid inequalities. 4 - Applying Stochastic Mixed-integer Programming to the Resiliency of Transportation Networks Amy Burton, PhD Student, Clemson University, Clemson, SC, United States, Akshay Gupte The decisions that dictate how funds are used to protect and increase the resilience of transportation networks should be made in a methodical, studied, and well-attested way. Focusing on bridges as critical links in transportation networks, we propose a stochastic mixed-integer program to minimize: (1) the investment costs of repairing and improving bridges and (2) time and travel costs incurred by motorists when old and damaged structures fail. Bender’s Decomposition is used to solve a convex formulation of the nonconvex mixed- integer nonlinear program. Computational results are presented for an example nine-node network as well as the benchmark Sioux Falls network. First Order Methods for Nonlinear Optimization Sponsored: Optimization/Linear and Conic Optimization Sponsored Session Chair: Negar Soheili Azad, University of Illinois-Chicago, Chicago, IL, 60607, United States 1 - In-face Frank-wolfe for Non-convex Optimization Paul Grigas, UC Berkeley, 4177 Etcheverry Hall, University of California, Berkeley, CA, 94720-1777, United States, Nathan Vermeersch The Frank-Wolfe method and its extensions are well-suited for delivering solutions with desirable structural properties, such as sparsity or low-rank structure. We adapt the methodology of in-face directions within the Frank-Wolfe method to the setting of non-convex optimization. We are particularly motivated by the application of this methodology to the training of neural networks with sparse and/or low-rank properties. We develop theoretical computational guarantees, and we complement these results with extensive numerical experiments. 2 - New Advances in Block Coordinate Descent Methods with Convergence to Second Order Stationary Solutions Songtao Lu, University of Minnesota Twin Cities, Room 4-174 200 Union St SE, Minneapolis, MN, United States The alternating gradient descent (AGD) is a simple but popular algorithm which has been applied to problems in optimization, machine learning, etc. In this talk, we show that a variant of AGD-type algorithms will not be trapped by ``bad’’ stationary solutions such as local maximum points. In particular, we consider a smooth unconstrained optimization problem, and propose a perturbed AGD (PA- GD) which converges (with high probability) to the set of second-order stationary solutions (SS2) with a global sublinear rate. To the best of our knowledge, this is the first alternating type algorithm which takes O(polylog(d)\epsilon^{7/3}) iterations to achieve SS2 with high probability. n SD05 North Bldg 122B

91